search

ge-xing / Diff-UNet

Diff-UNet: A Diffusion Embedded Network for Volumetric Segmentation. (using diffusion for 3D medical image segmentation)
Apache License 2.0
141 stars 20 forks source link

How is the embed_model trained #4

Closed SkeletonOne closed 1 year ago

SkeletonOne commented 1 year ago

Hi,

I notice that the optimizer is defined only for the denoising unet model here https://github.com/ge-xing/Diff-UNet/blob/d954bdced66fd245bb20123bfd053a1ae382ba41/BTCV/train.py#L88, and I cannot find the optimizer for the embed_model (feature encoder in the paper). Does it mean the FE is just randomly initialized and not trained? Or are there something I am missing? Thanks.

ge-xing commented 1 year ago

Self.model is the instance of Diff-Uet, it contains the embed_model.

SkeletonOne commented 1 year ago

Got it. Thanks!